You only need a couple GB to make a game look as good as possible given the compute resources available today. The only reason some games use more is due solely to the laziness of the devs. 2GB is enough to store all visual information within a normal first person view vision range. If you were to switch to a scope, it is perfectly acceptable for there to be a 100mS lag if additional textures have to be loaded. Games wouldnt even be realistic if you could just toggle a scope and instantly see at 50x magnification anyway. So the lag induced by the memory filling synergizes with the desire for realism. Aside from scoping, there is no need to load more ultra-hi-res textures than what the character can see or travel to in a set amount of time. Unless we're talking about superhuman type characters that can zoom around massive areas quickly and the devs want all that data to be instantly accessible. But really, how many games are like that? That basically only happens in driving games, which are usually very good at dynamic loading. Even in driving games, the frame of reference moves around at somewhat predictable speeds which makes dynamic loading relatively easy. Memory bandwidth is just so much more important that actual capacity. Dynamic loading will occur very frequently regardless of how much VRAM you have, because it is never enough to hold everything.
How many games can you walk up to things and
not see them get fuzzy? So far I'm at 0, which, as far as I'm concerned, completely destroys your argument. Textures will not be detailed enough until every object that's right up against the "camera" has no obvious stretching of textures on its surfaces. There's no need to be zooming. It happens with typical 60-90 degree FOVs.
As well, if you have a free camera, be it third or first person, you can generally make a 360 in 100-200ms, with no disorientation, and the game should be able to remain detailed and fluid throughout that time. That should generally be plenty of time to grab a texture and put it in VRAM, too,
if the game can use system RAM to cache it, in lieu of huge VRAM amounts, and/or can assume it will be loaded from an SSD (assuming only a few textures actually aren't already in VRAM, since it's the same space, and most of it will be needed at any angle).
Today, we still don't have anything even close to that, even in games that genuinely do look amazing, like Ryse (if only its gameplay could match its graphics--it's like Crysis, in that respect). A texture that appears 2x as detailed, or 2x less fuzzy, takes just under 3x more space the combination of less detailed textures, so the VRAM usage, if not managed by necessary LOD at the moment, would balloon, not merely bump up a tad. Instead of using LODs and high res textures to give that for close objects, but still drop down at distances, we get things that look decent at a perceived 10-20ft, and then that fuzzy wet marker look, closer.
What you see at a distance is of curse a different matter, and should be more aggressively managed, IMO, to free up some of that VRAM that isn't needed, but we should have the resources today to remove perceived pop-in. Maybe with decent [V]RAM in the new consoles, engines over the next few years will take care of that.
Facebook
Twitter